def get_weight(dt, isotope='18F', a=1, t=None, r=None, tac=None):
if isotope == '18F':
half_life = 20.39 * 60
elif isotope == '12C':
half_life = 109.77 * 60
else:
print('not known yet')
decay_const = np.log(2) / half_life
tdur = kt.dt2tdur(dt)
dcf = decay_const * tdur / (np.exp(-decay_const * dt[0, ]) -
np.exp(-decay_const * dt[1, ]))
n_model = 5
w_m = [None] * n_model
i = 0
w_m[i] = a * dcf * dcf * t / tdur / tdur
i += 1
w_m[i] = a * dcf * dcf * (t + 2 * r) / tdur / tdur
i += 1
w_m[i] = a
i += 1
w_m[i] = a * dcf * dcf
i += 1
w_m[i] = a * tac * tdur * dcf
for i in range(n_model):
w_m[i] = 1 / w_m[i]
return w_m
def mrtm_k2p(tac, dt, inputf1, k2p,
linear_phase_start, linear_phase_end, fig, fig_name):
if linear_phase_start is None:
linear_phase_start = 0
if linear_phase_end is None:
linear_phase_end = np.amax(dt)
mft = kt.dt2mft(dt)
mft = np.append(0, mft)
dt_new = np.array([mft[:-1], mft[1:]])
tdur = kt.dt2tdur(dt_new)
input_dt = kt.int2dt(inputf1,dt)
tac = np.append(0, tac)
input_dt = np.append(0, input_dt)
tac_cum = np.cumsum((tac[:-1] + tac[1:]) / 2 * tdur)
input_cum = np.cumsum((input_dt[:-1] + input_dt[1:]) / 2 * tdur)
tac = tac[1:]
input_dt = input_dt[1:]
yy = tac
xx = np.column_stack((input_cum + 1 / k2p * input_dt, tac_cum))
tt = np.logical_and(mft > linear_phase_start, mft < linear_phase_end)
tt = tt[1:]
mft = mft[1:]
reg = LinearRegression(fit_intercept=False).fit(xx[tt, ], yy[tt])
bp = - reg.coef_[0]/reg.coef_[1] - 1
yyf = reg.predict(xx)
if fig:
plt.cla()
plt.plot(mft, yy, '.')
plt.plot(mft, yyf, 'r')
if fig_name is not None:
plt.savefig(fig_name)
plt.show()
kps = {'bp': bp}
return kps
def mrtm(tac, dt, inputf1, linear_phase_start, linear_phase_end, fig):
if linear_phase_start is None:
linear_phase_start = 0
if linear_phase_end is None:
linear_phase_end = np.amax(dt)
mft = kt.dt2mft(dt)
mft = np.append(0, mft)
dt_new = np.array([mft[:-1], mft[1:]])
tdur = kt.dt2tdur(dt_new)
input_dt = kt.int2dt(inputf1, dt)
tac = np.append(0, tac)
input_dt = np.append(0, input_dt)
tac_cum = np.cumsum((tac[:-1] + tac[1:]) / 2 * tdur)
input_cum = np.cumsum((input_dt[:-1] + input_dt[1:]) / 2 * tdur)
tac = tac[1:]
input_dt = input_dt[1:]
yy = tac
xx = np.column_stack((input_cum, tac_cum, input_dt))
tt = np.logical_and(mft > linear_phase_start, mft < linear_phase_end)
tt = tt[1:]
mft = mft[1:]
reg = LinearRegression(fit_intercept=False).fit(xx[tt, ], yy[tt])
bp = -reg.coef_[0] / reg.coef_[1] - 1
k2p = reg.coef_[0] / reg.coef_[2]
# for 1 TC
r1 = reg.coef_[2]
k2 = -reg.coef_[1]
yyf = reg.predict(xx)
if fig:
plt.plot(mft, yy, '.')
plt.plot(mft, yyf, 'r')
plt.show()
kps = {'bp': bp, 'k2p': k2p, 'r1': r1, 'k2': k2}
return kps
def logan_ref_k2p(tac, dt, inputf1, k2p, linear_phase_start, linear_phase_end,
fig):
if linear_phase_start is None:
linear_phase_start = 0
if linear_phase_end is None:
linear_phase_end = np.amax(dt)
mft = kt.dt2mft(dt)
mft = np.append(0, mft)
dt_new = np.array([mft[:-1], mft[1:]])
tdur = kt.dt2tdur(dt_new)
input_dt = kt.int2dt(inputf1, dt)
tac = np.append(0, tac)
input_dt = np.append(0, input_dt)
tac_cum = np.cumsum((tac[:-1] + tac[1:]) / 2 * tdur)
input_cum = np.cumsum((input_dt[:-1] + input_dt[1:]) / 2 * tdur)
tac = tac[1:]
input_dt = input_dt[1:]
yy = np.zeros(tac.shape)
xx = np.zeros(tac.shape)
mask = tac.nonzero()
yy[mask] = tac_cum[mask] / tac[mask]
xx[mask] = (input_cum[mask] + input_dt[mask] / k2p) / tac[mask]
tt = np.logical_and(mft > linear_phase_start, mft < linear_phase_end)
tt = tt[1:]
dvr, inter, _, _, _ = linregress(xx[tt], yy[tt])
bp = dvr - 1
yyf = dvr * xx + inter
if fig:
plt.plot(xx, yy, '.')
plt.plot(xx[tt], yyf[tt], 'r')
plt.show()
kps = {'bp': bp}
return kps
def image_to_suvr_with_reference_tac(pet_image, dt, reference_regional_tac,
selected_frame_index):
tdur = dt2tdur(dt)
image = np.multiply(pet_image[:, :, :, selected_frame_index],
tdur[selected_frame_index])
ref = np.multiply(reference_regional_tac[selected_frame_index],
tdur[selected_frame_index])
image_suvr = np.sum(image, axis=-1) / np.sum(ref)
return image_suvr
def mrtm(tac, dt, inputf1, linear_phase_start, linear_phase_end, fig):
if linear_phase_start is None:
linear_phase_start = 0
if linear_phase_end is None:
linear_phase_end = np.amax(dt)
# fill the coffee break gap
if kt.dt_has_gaps(dt):
tac, dt = kt.tac_dt_fill_coffee_break(tac, dt)
mft = kt.dt2mft(dt)
mft = np.append(0, mft)
dt_new = np.array([mft[:-1], mft[1:]])
tdur = kt.dt2tdur(dt_new)
input_dt = kt.int2dt(inputf1, dt)
tac = np.append(0, tac)
input_dt = np.append(0, input_dt)
# set negative values to zero
tac[tac < 0] = 0.0
input_dt[input_dt < 0] = 0.0
tac_cum = np.cumsum((tac[:-1] + tac[1:]) / 2 * tdur)
input_cum = np.cumsum((input_dt[:-1] + input_dt[1:]) / 2 * tdur)
tac = tac[1:]
input_dt = input_dt[1:]
yy = tac
xx = np.column_stack((input_cum, tac_cum, input_dt))
# find tt for the linear phase
tt = np.logical_and(mft >= linear_phase_start, mft <= linear_phase_end)
tt = tt[1:]
# select tt for tac > 0
tt = np.logical_and(tt, tac > 0)
# select tt for xx < inf, yy < inf
infinf = 1e10
tt = np.logical_and(tt, np.all(xx < infinf, axis=-1))
tt = np.logical_and(tt, yy < infinf)
mft = mft[1:]
reg = LinearRegression(fit_intercept=False).fit(xx[tt, ], yy[tt])
bp = -reg.coef_[0] / reg.coef_[1] - 1
k2p = reg.coef_[0] / reg.coef_[2]
# for 1 TC
r1 = reg.coef_[2]
k2 = -reg.coef_[1]
yyf = reg.predict(xx)
if fig:
plt.plot(mft, yy, '.')
plt.plot(mft, yyf, 'r')
plt.show()
kps = {'bp': bp, 'k2p': k2p, 'r1': r1, 'k2': k2}
return kps
def logan_ref_k2p(tac, dt, inputf1, k2p, linear_phase_start, linear_phase_end,
fig):
if linear_phase_start is None:
linear_phase_start = 0
if linear_phase_end is None:
linear_phase_end = np.amax(dt)
# fill the coffee break gap
if kt.dt_has_gaps(dt):
tac, dt = kt.tac_dt_fill_coffee_break(tac, dt)
mft = kt.dt2mft(dt)
mft = np.append(0, mft)
dt_new = np.array([mft[:-1], mft[1:]])
tdur = kt.dt2tdur(dt_new)
input_dt = kt.int2dt(inputf1, dt)
tac = np.append(0, tac)
input_dt = np.append(0, input_dt)
# set negative values to zero
tac[tac < 0] = 0.0
input_dt[input_dt < 0] = 0.0
tac_cum = np.cumsum((tac[:-1] + tac[1:]) / 2 * tdur)
input_cum = np.cumsum((input_dt[:-1] + input_dt[1:]) / 2 * tdur)
tac = tac[1:]
input_dt = input_dt[1:]
yy = tac_cum / (tac + 0.0000000000000001)
xx = (input_cum + input_dt / k2p) / (tac + 0.0000000000000001)
# find tt for the linear phase
tt = np.logical_and(mft >= linear_phase_start, mft <= linear_phase_end)
tt = tt[1:]
# select tt for tac > 0
tt = np.logical_and(tt, tac > 0)
# select tt for xx < inf, yy < inf
infinf = 1e10
tt = np.logical_and(tt, xx < infinf)
tt = np.logical_and(tt, yy < infinf)
# do linear regression with selected tt
xx = xx[tt]
yy = yy[tt]
dvr, inter, _, _, _ = linregress(xx, yy)
bp = dvr - 1
yyf = dvr * xx + inter
if fig:
plt.plot(xx, yy, '.')
plt.plot(xx[tt], yyf[tt], 'r')
plt.show()
kps = {'bp': bp}
return kps
tac = np.array([
0.0612937319754616, 2.56733416202755, 13.2302628562565, 17.5212376697891,
19.8554469827973, 21.2888518029975, 22.3164453083215, 21.9106339934451,
21.6976991361882, 21.2760967663021, 21.0320027737135, 20.3356078483447,
19.7789261620624, 18.7016942272702, 17.5381378584246, 16.6955628809652,
15.8305514461795, 15.1466577726934, 14.2137260024906, 13.7598553636470,
12.9989332998008, 11.8293518915139, 10.3648913297944, 8.93110113265639,
7.50503727314817, 6.49229487908344, 5.83027369667342, 5.35683342894270,
4.94744092837665, 4.79454967790696, 4.49907835991146
])
tac = TAC(tac, dt)
ref = Ref(ref, dt)
ref.interp_1()
w = dt2tdur(tac.dt)
w = w / np.amax(w)
beta_lim = [0.0100000 / 60, 0.300000 / 60]
n_beta = 40
b = basis.make_basis(ref.inputf1, dt, beta_lim=beta_lim, n_beta=n_beta, w=w)
# provide all user inputs in one dict here and later 'get_model_inputs' will select the needed ones
user_inputs = {
'dt': dt,
'inputf1': ref.inputf1,
'w': w,
'r1': 0.905,
'k2p': 0.00025,
'beta_lim': beta_lim,
'n_beta': n_beta,
'b': b,
def logan_ref(tac, dt, inputf1, linear_phase_start, linear_phase_end, fig):
if linear_phase_start is None:
linear_phase_start = 0
if linear_phase_end is None:
linear_phase_end = np.amax(dt)
# fill the coffee break gap
if kt.dt_has_gaps(dt):
tac, dt = kt.tac_dt_fill_coffee_break(tac, dt)
mft = kt.dt2mft(dt)
mft = np.append(0, mft)
dt_new = np.array([mft[:-1], mft[1:]])
tdur = kt.dt2tdur(dt_new)
input_dt = kt.int2dt(inputf1, dt)
tac = np.append(0, tac)
input_dt = np.append(0, input_dt)
# set negative values to zero
tac[tac < 0] = 0.0
input_dt[input_dt < 0] = 0.0
# calculate integration
tac_cum = np.cumsum((tac[:-1] + tac[1:]) / 2 * tdur)
input_cum = np.cumsum((input_dt[:-1] + input_dt[1:]) / 2 * tdur)
# tac_cum and input_cum are calculated in such a way to match tac
# # # debug
# file_path = '/Users/Himiko/data/amsterdam_data_pib/transfer_162391_files_89f2cba1/'
# savetxt(file_path + 'tac_cum.csv', tac_cum)
# savetxt(file_path + 'input_cum.csv', input_cum)
#
# tac1 = kt.interpt1(mft, tac, dt)
# input1 = kt.interpt1(mft, input_dt, dt)
# tac_cum1 = np.cumsum(tac1)
# input_cum1 = np.cumsum(input1)
# tac_cum1_mft = tac_cum1[mft.astype(int)]
# input_cum1_mft = input_cum1[mft.astype(int)]
# savetxt(file_path + 'tac_cum1_mft.csv', tac_cum1_mft)
# savetxt(file_path + 'input_cum1_mft.csv', input_cum1_mft)
# # # debug
tac = tac[1:]
input_dt = input_dt[1:]
# yy = np.zeros(tac.shape)
# xx = np.zeros(tac.shape)
# mask = tac.nonzero()
# yy[mask] = tac_cum[mask] / tac[mask]
# xx[mask] = input_cum[mask] / tac[mask]
yy = tac_cum / (tac + 0.0000000000000001) # ADDED BY MY 20210616
xx = input_cum / (tac + 0.0000000000000001) # ADDED BY MY 20210616
# find tt for the linear phase
tt = np.logical_and(mft >= linear_phase_start, mft <= linear_phase_end)
tt = tt[1:]
# select tt for tac > 0
tt = np.logical_and(tt, tac > 0)
# select tt for xx < inf, yy < inf
infinf = 1e10
tt = np.logical_and(tt, xx < infinf)
tt = np.logical_and(tt, yy < infinf)
# do linear regression with selected tt
xx = xx[tt]
yy = yy[tt]
dvr, inter, _, _, _ = linregress(xx, yy)
bp = dvr - 1
yyf = dvr * xx + inter
if fig:
plt.plot(xx, yy, '.')
plt.plot(xx, yyf, 'r')
plt.show()
kps = {'bp': bp}
return kps